YouGov’s weekly Sunday Times poll is now up here. Topline voting intention are CON 34%, LAB 34%, LDEM 7%, UKIP 14%, GRN 5%.

Most of the survey was made up of questions about the budget and government spending. If George Osborne has money to spend in the budget 44% would prefer it goes on public services, 25% on tax cuts, 20% on the deficit. In general people would like to see any spending focused up helping low paid people in work (59%), followed by people looking for work (31%), small businesses (25%) and homebuyers (25%). People saving for their retirement, incidentally, comes bottom.

On specific measures most of those YouGov tested got the thumbs up – the most widespread approval was for increasing the personal tax allowance again (83%), limiting child benefit to three children (73%) and raising the NI threshold (71%). Letting people buy back annuities they bought when they were compulsory gets low support, but mainly because of a very high don’t know (I expect people simply don’t understand the change). The only measure that was actually opposed by more people than supported it was cutting taxes on alcohol (33% would support, 50% would oppose).

Moving onto government spending in general the areas people would most like to see protected from government cuts are the NHS (79%), education (50%) and policing (35%). The areas people most wanted to see cut were overseas aid (66%), welfare benefits (36%) and environment and climate change (29%). As I discussed in the weekly round up, defence and welfare were unusual in being issues that had both significant numbers of people wanting to prioritise them for cuts and significant numbers of people wanting to protect them from cuts.

Asked specifically about whether the government should commit to 2% of GDP spending on defence, 52% think they should, 27% that they should not. Asked the equivalent question about overseas aid only 24% think the government should commit to the 0.7% target, 59% think they should not. On Trident, 31% think it should be replaced with an equally robust system, 29% replaced with a cheaper system, 24% scrapped completely.

Outside of Scotland itself, the idea of the SNP being in a position of influence at Westminster is seen negatively – 63% think it would be a bad thing if they held the balance of power in Westminster, 64% think it would be bad thing if they were involved in a coalition. Overall 53% of people think that Labour should rule out doing a deal with the SNP, but this is largely made up of Labour’s opponents, their own supporters are far more split over the idea. If there was a choice between a minority Labour government or an SNP/Lab coalition with a majority, Labour voters would be evenly divided but if the alternative was another Tory government Labour voters would back a deal with the SNP by 6 to 1.


527 Responses to “YouGov/Sunday Times – CON 34, LAB 34, LD 7, UKIP 14, GRN 5”

1 2 3 4 11
  1. Will someone please tell me which pollsters now prompt UKIP and which do not?

  2. Latest cover up -some very bright spark is running a “how many seats will we get “competition at libdem conf.Cleggies want results hidden from hacks.

    http://www.theguardian.com/politics/2015/mar/15/lib-dem-members-run-sweepstake-on-general-election-seat-retention

  3. Simon,

    Another nudge to get me to do the methodology summary of all the pollsters that I keep meaning to do!

    Companies that do are Populus, Ashcroft, ComRes, YouGov and Survation

    Companies that do not are Ipsos MORI, ICM, Opinium

  4. “I’ve noticed that most of the predictions have been showing the Tories ahead on seats even though the polls have been roughly equal for ages. Yet. Yet traditionally Tories had to be several percent ahead to get the most seats. That goes a long way to explain the change.”

    Another piece of craziness, if i may be so candid….

    the biggest driver in this election is the conservative to labour swing in england which is entirely independent of what happens in scotland.

    the swings required in england for tory seats to fall into labour hands at the next election have not changed. the boundaries are exactly the same as they were in 2010…nothing that happens in scotland has made tory held marginals any less or more likely to fall to labour…the arithmetic has NOT changed….

    Now most of these projections, predictions are factoring quite a bit of swingback. Some of the models having predicted, frankly, barmy outcomes are becoming more sensible…viz. Fisher…

    some of them electionforecast and others are veering on the crazy. again common sense breaks in. To believe that the tories will win 295 seats at the election in just over 7 weeks time…you have to believe that the blues will gain something like 13 lib seats and only lose 20 to labour, and none to ukip….I notice having shown the tories on 295, hanretty’s “model” now shows them on 286…these “models” are an embarrassment….

    The Con to labour swing in England, as i have repeated ad nauseam, is the key. the tories beat labour by 11% in England in 2010. I believe they need to beat labour by 7% in England, at least, to have a shot at government…

    None of the data we have seen in the last 3 months has suggested anything like this, but of course, either a) the polls are plain wrong or b) there could be significant swingback, say 2%, to the tories in the next 7 weeks.

  5. Anthony,

    Thank you.

    How do MORI, ICM and Opinium calculate the UKIP percentage?

  6. chrislane1945

    Had a look at the May2015 site linked to in your tweet.

    Presumably their “If LAB win few Scottish seats FPTP no longer favours them. CON win if tie in polls” means that the supposed structural advantage that Labour was supposed to have was simply based on Scotland returning lots of Lab MPs.

    Therefore, Labour doesn’t have a structural advantage in England.

    Those, still pointlessly using GB numbers and archaic assumptions, now discover that they are rather useless.

    Or is there another explanation?

  7. Surprised that the total of people who wanted to spend more on in work benefits and cut welfare did not add up to significantly more than 100%.

    On ‘vice duties’ (taxes on cigarettes, alcohol and the like), if a chewing gum tax isn’t brought in soon I’ll seriously consider subsidising regular weekend breaks in Europe by paying lower rates of duty on things I currently buy in the UK.

  8. @ Peter Crawford

    You are perfectly right. Methodologically these models are extremely dubious.

    However, there are no “events” (the last election was 5 years ago), so the models cannot be improved (For a particular purpose I have had to work on Bayesian stat, hence the use of their thinking in some of my posts), or the changes are completely arbitrary.

    It seems to me that most models are built around the 2010 elections with modifiers from previous elections, and other arbitrary injections.

    The only reasonable data we have are the voting intentions. The trouble is the FPTP, so if we don’t want to use UNS, we resort to exactly the same thing as the models.

    The other possibility is your suggestion: building up a logical tree, and measure it against our current understanding. While I fully agree with this, the trouble is that the data is not detailed enough, so it would require a massive swing to invalidate the logical connections in the tree. In absence of the data, the verification is just common sense, even if I agree with you.

    A secondary from this is that the election will verify or reject your model, but will not say anything about the next elections (2020).

  9. Simon – Opinium (who are online) do what other online companies used to do, people can click “other” to get a list of further options which includes UKIP.

    ICM and MORI are telephone polls, so people can just say “UKIP” when asked who they’ll vote for and ICM and MORI record it.

  10. @Oldnat – “Presumably their “If LAB win few Scottish seats FPTP no longer favours them. CON win if tie in polls” means that the supposed structural advantage that Labour was supposed to have was simply based on Scotland returning lots of Lab MPs.

    Therefore, Labour doesn’t have a structural advantage in England.”

    I think there is another obvious explanation.

    Once the SNP get past 40% in Scotland, FPTP gives them a clear structural advantage – despite getting nowhere near a total vote share of 80%, they will win the vast majority of the seats.

    In England, Labour still retains a substantial advantage from FPTP, but this is neutralized by their disadvantage in Scotland.

    I would have thought understanding this would be fairly straightforward, although your central point stands – the use of a single GB wide seat allocations is a bit pointless, as there is no longer a single direction in the way FPTP favours specific parties.

  11. Alec

    Thanks. I’m watching the Cup Final, so I’m not concentrating too much on the Maths!

    But, if Lab are ahead of Con (OK not by a huge amount) in Scotland, that adds a bit to the Lab side of Lab v Con VI in GB.

    How then can “CON win if tie in [GB] polls” if Lab have a structural advantage in England?

  12. @ Laszlo

    I built a model of the proposed boundary changes which did not come to pass; also a model to forecast AV outcomes, again the AV proposal did not pass. I was planning to build a model for the GE but the Scottish referendum campaign used up all my spare time.

    I would like to tell you about the idea for the model. I think you will understand the idea, even though you may not approve of it! It was to be a what-if calculator with probable events & probable range of outcomes.

    Thus a scenario could be cascaded through the model. e.g. What if the chancellor’s final budget is neutral; or the best-ever budget for a chancellor; or the worst ever? This would give a range of outcomes.

    To forecast it, the probability of each outcome would need to be calculated based on history/ patterns for that chancellor/ ‘gut feel’. So it would not be ‘just the facts’. There would be an element of political judgement.

    It would have the capability to forecast using multiple events as the input criteria.

    The ‘dream’ model is to have the likely impact roll out for each constituency, based on its voting history & demographics!

    I may build it for 2020 (assuming the fixed term parliament continues).

  13. Anthony,

    I note Opinium are online and ICM and MORI are telephone.

    Are Ipos, Populus, Ashcroft, ComRes, YouGov and Survation online or telephone?

    Is there anywhere on this website giving the info I have requested (i.e. which pollsters prompt UKIP and which polling methods are used by the pollsters)? If not, I think it would be useful.

    [There will be soon – I keep meaning to write a nice post summarising them all, it’s actually about half done, so hopefully this week – AW]

  14. @ Chris Hornet,

    if a chewing gum tax isn’t brought in soon

    Oi!

  15. The simple question is say we get 35/35 in England and Wales, what are the likely seat numbers? Labour must have a chunky lead with all of those Welsh MP’s and you don’t need many votes to win a seat there.

  16. Rob Ford of Revolt on the Right fame just published an article in the-paper-which-must-not-be-linked claiming that Labour still have an overall efficiency advantage in their vote distribution even with the rise of the SNP, so there’s some disagreement over this in academia.

    I’d link it but a) It Is Forbidden and b) there are no numbers in it, so we have no way of knowing how he arrived at this conclusion and there’s nothing for us to debate. He’s generally pretty credible, though.

  17. Another point. Do YouGov poll in Northern Ireland. In the details of their poll there is no mention of the Northern Ireland parties.

    [No. All the regular polls are Great Britain only – AW]

  18. @CMJ and others

    YouGov polls DO favour Labour more later in the week!

    I have extended your ANOVA analysis of polls since Jan 1, but tweaked in a way to reduce variability. The problem with your original ANOVA analysis is that both Tory and Labour VIs have been jumping around between Jan 1 and now. If you get a good week for the Tories (like last week) then each of the VIs will be up a bit. Similarly, for a good Labour week, the Labour vs Tory margins will rise a little. If you set out to compare Monday polls with Tuesday’s, Wednesdays etc, then the good and bad weeks will contribute to increased variability of the figures making up the group’s for ‘Monday’, ‘Tuesday’ etc.

    To get round this, for the five YouGov polls of each week I derived a single figure to capture Labour’s margin drop from Monday to Friday. I then submitted the ten figures for this year to a test to see whether they differ significantly from zero.

    More specifically, the figure I used was the linear decline in Labour’s margin over the Tories from Monday to Firday. So, to use an extreme and unrealistic illustration if Labour were three points behind the Tories on Monday, two behind for Tuesday’s poll, one behind on Wednesday, tied on Thursday and one ahead for the Friday poll reported in the Sunday Times…then this would be recorded as an (unrealistically large) one unit per day improvement for Labour over the week. If the margins were the same for each of the five days then the ‘improvement, would be recorded as zero. A drift in the opposite direction would be recorded as a negative score.

    Over recent weeks, UKPR folk have been speculating about whether the beginnings of weeks are bad for Labour, with things getting better for them over the week. In terms of the analysis above, this would show up as a series of increased Labour margins across the successive weeks of the year.

    Much to my surprise, the analysis shows that the IS statistical support for this trend. The relevant p-value was 0.033.

    As with all statistical findings, you could get such an effect by chance even if there are no genuine variations across the different days of the week. (the p-value indicates that the chance of this happening is about one in thirty).

    Alternatively, there may indeed be something about weekly shifts in mood that cause members of the YouGov panel to respond in slightly different ways as each week proceeds.

    On average, since Jan 1 Labour’s margins has improved by about 1% between the beginning and the end of the week.

    Distinctly odd!

  19. I’m going to repost this for those who missed it last time, because it directly addresses the “structural advantage” debate and moreover there are some figures and graphs

    http://www.economist.com/news/britain/21643144-one-eternal-facts-british-politics-no-longer-true-labour-surge

  20. Spearmint

    I understand the argument that a smaller % of Labour voters in safe Labour seats actually turn out to vote compared with Tories in safe Tory seats.

    But is that tendency replicated in the polls?

  21. @ Old Nat,

    Well, presumably the polls wouldn’t be predictive if it wasn’t, and they are (reasonably so, anyway), so it must be factored in somehow.

    Likelihood to vote adjustments or differential response rates (if you’re an apathetic voter in a safe Labour seat, you might not respond to a YouGov poll any more than you would to an election) are two ways you could represent this tendency in the polls, but this is really a question that requires Anthony’s expertise.

  22. @ Unicorn & Chrislane 1945

    I looked at the polls from March 6th to 13th and crudely compared them to the GE 2010 results for those regions, just for purposes of considering the statistical inferences. There are a number of variables to consider:

    1. The fact that neither LD or UKIP have yet nominated the number of candidates they did last time whereas Green are already up 172 on last time and will likely have a full slate in London, the Southeast and Southwest. Just ten candidates short right now.

    2. In 1993 in Canada I realized that there were distinct election “sub-national” races going on that “national” polls were not catching, like determining last night that over the last week the swing to Labour in the North was supposedly barely .5%.

    This would confirm my observation that UKIP is likely pulling more votes from Labour in the North and Conservative in the South. In 1975 I worked on a constituency campaign in which the NDP (Labour) in a one on one race with Socred (Conservative) won. We were ecstatic, until on election night we watched our Party as a whole go down to defeat, so you have to keep your eye on the macro and micro pictures simultaneously.

    3. I am extremely skeptical of pollsters who are effectively upweighting parties to 2010 GE results, such as seeing the raw number of respondents for LD at say 27% of 2010 be increased to a weighted number of 31%. This tells me that if the pollsters are wrong with this weighting that the situation for LD is even more dire.

    Just think about it for a moment, in the North and Midlands last week three pollsters found Green ahead of LD, even with their current weighting to 2010.

    4. Even with Comres factored in, I simply do not believe Green are at 4% based on the increase in membership and activity on the ground, they are still ahead of LD in the North and Midlands, within .7% in the Southeast and supposedly within 1.3% in Wales – Southwest.

    Let’s just think about that for a minute. LD were second in the Southwest with 34.7% in 2010. I am constantly being challenged on my idea that the 2014 EU results should be factored in along with GE 2010 results, but how else do you accurately shape feedback from the changes that are happening within the electorate into your polling model?

    YouGov have begun to factor in the 2014 Scottish Referendum Results, why not the 2014 EU results as well?

    5. I had LD and Green on 7.3% each in the first week of February and agree that in terms of the polling results that LD have pulled ahead of Green since, but just found that they themselves are starting to show signs of sinking below that first week of February level.

    I cannot absolutely confirm it because I have not measured it, but from a pure eyeball glance I believe LD are dropping below their EU support level in Scotland and the North, and may be starting to do so in the Midlands. So LD could experience some kind of “swingback” in certain regions of England, but not in others.

    6. Following the sub-regional text in this UK GE is going to be how one determines who wins what in the election, and is how I was successfully able to predict the near absolute wipeout of the Progressive Conservatives in Canada in 1993 and the fact that the Green were going to win a seat in the 2014 EU in the Southwest.

    5. Now to seat numbers. Looking at the 1922, 1970 and 1974 UK GE, and 1993 and 1997 Canadian I question the veracity of the seat numbers being produced by election forecast models used in this UK election.

    They simply do not line up with the results that have occurred under FPTP. Further Ashcroft polling, particularly, is showing that incumbency is not a show stopper in terms of loss of seats. Let’s take Twickenham, for example:

    i. LD win in 2010 with 54.4%

    ii. A 67.2% drop in LD support in the southeast nets a result of 17.8%

    iii. classical Canadian political science texts talk about the candidate factor being as high as 33% of the reason why a person votes a certain way, so plugging that in I can get Vince Cable to 23.7%, if the polling figures are accurate.

    7. That means that in the Southwest I can get LD incumbents to roughly 41% of what they achieved in 2010, which is the highest for them anywhere in GB currently.

    Of course all this changes if the polling numbers move, which they will, but I just wanted to crudely plug in last week’s polling numbers and see what I got.

    This is, after all supposed to be a site that discusses polling numbers and presumably their implications, not a place to speculate about what might or might not happen, as I for one am not a soothsayer.

    I can, however, look at a set of polling numbers and say based on these numbers at this time, we can potentially expect x, y and z to happen under FPTP.

    8. At 20% or higher you can expect any political party to start picking up seats within a region based on a whole range of demographic and socio-economic variables, but at 9% or less you have to wonder where a political party can begin to save any seat including centuries old bastions.

  23. On second thought, it occurs to me this could potentially screw up likelihood-to-vote weighings as far as seat prediction is concerned. You’d get the right national VIs (which of course is how the pollsters actually measure their accuracy, so they’re not bothered), but when it came to the seat prediction all the apathetic Labour voters in safe seats would weigh down the Labour VI with their lousy likelihood scores, and so the marginals would be skewed Labour-ward relative to the national polling.

    Maybe Tory incumbency balances it out? And most of the pollsters don’t do likelihood adjustments (yet- when does YouGov start?), so it probably isn’t distorting up the poll-of-polls too badly. But if you were trying to calculate based on just eg. Ipsos Mori, you could be in trouble.

  24. Spearmint

    It was really a rhetorical question. It would be good to get Anthony’s insight – as well as what (if anything) the May2015 tweet meant.

  25. @Unicorn

    I get suspicious when the raw data is tweaked.

    A straight forward analysis of recorded values shows no link. If by reducing variability.

    Techniques to rework the data variability first doesn’t feel right at all.

  26. I feel the variability of the data removed is part of the data you need to incorporate into the test.

  27. TwoNil, TwoNil, TwoNil…….

    One down could be two to go!

    Peter.

  28. Even if Labour do have statistically higher polling later in the week, it would just be a spurious correlation.

  29. Interesting FT article on the Budget, in particular what may result from the Conservative/Lib Dem compromise

    http://www.ft.com/cms/s/0/96da772c-cb20-11e4-8ad9-00144feab7de.html#axzz3UTTNCk1D

  30. If Unicorn is right about Yougov polls favouring the Tories early in the week and Labour at the end of the week, I wonder if Labour will benefit from the election being on a Thursday?!

  31. @Politicainado

    In a word no.

    Does anyone really believe that a voter would vote Conservative on Monday, and some how change to Labour on Thurday – becuase it’s Thursday?

  32. Jeff, no, but if Unicorn is right (and I’ve observed the trend), an explanation is required. Perhaps don’t knows feel more warm towards Labour as the weekend nears after a hard week.

  33. @Unicorn – not sure how this works, but my initial point on this wasn’t based on the lead, but the presence or absence of a Con lead. I think I’m correct in saying of something like 11 YG con leads, 9 have been on Mon/Tues (this year). The Y/N Con lead question could be more significant ?

  34. @CMJ

    Always right to be suspicious.

    Let me try a different way of explaining what I have done. Imagine a situation in which there is a small, regular real decline over the week of 0.2 units a day. Suppose also that average scores are going up by (say) 0.5 per week.

    So the data sequence might go something like this: 1.0, 0.8, 0.6, 0.4, 0.2 for Week 1 following by 1.5, 1.3, 1.1, 0.9 and 0.7 for Week 2 and so on until Week 10 produces the sequence: 5.5, 5.3, 5.1, 4.9 and (for the Friday score) 4.7.

    On these assumptions the Monday scores will range from 1.0 to 5.5 and the Friday scores from 0.2 to 4.7. With this amount of variation it would probably be impossible to show that Monday scores are higher than those for Fridays.

    But – by hypothesis – much of the variation comes from overarching changes that are nothing to do with the day of the week.

    To be able to detect the effect of interest (the effect of day of week) you need to be able to partial out the changes that would otherwise be distorting the weekly patterns.

    To do this, the approach I have adopted is to work out the size of the linear drop each week, and then to test whether the ten weekly drop figures differ from zero. If there is a drop over the week whatever the starting point then I don’t think this represents gratuitous tweaking.

    If you preferred you could derive weekly scores for the difference between – on the one hand – the average Labour-over-Tory margin for Mondays and Tuesdays and – on the other – the average margin for Thursdays and Fridays.

    In effect you are sampling the drop on a weekly basis and then later testing whether these sampled values differ from zero.

    Do you really consider this a suspicious way of handling the data if you are trying to test whether there is a drop over the week?

  35. Only possibly explanation if the phenomena is real would be sample bias based on day of the week. Unlikely.

  36. AW himself has looked at this several times and come to to the conclusion there is no proven links to days of the week.

    As regards to what you have observed, how do you know that what has been observed isn’t just coincidence, and not a substantial trend or pattern?

    You use data analysis to confirm if your observation really is ‘proven’. I used the same data in the basic form, there get a different result.

    @Unicorn took steps to reduce variability in the data before applying a test. So in a sense, different things were measured – one test raw data and one test measured manipulated data.

    Who is right?

    As the two tests measured different things, you can’t really compare the results.

  37. @Alec

    My analysis used rather more information than whether there was a simple Labour-over-Tory lead. I subtracted the Tory from the Labour VI in order to use the size of the lead in my calculations. If the YouGov polls are slightly kinder to Laber at the beginning of the week, then this should show up in the form of bigger leads on average near the beginning of the week.

    You were talking about more leads. But it is a different way of talking about the same thing.

  38. Peter Crawford
    For what its worth I don’t subscribe to the “Gag Reflex” theory and yes the likelihood is simple margin of error etc is responsible BUT it must be concerning for the Tories that with election day fast approaching they seem incapable of opening up a consistent lead of any size. That could obviously change but if I were a Tory I’d be getting quite worried.

  39. I will think about more when:

    a) I’ve cooked tea
    b) I’ve got the kids to bed

    I’ve enough grief this afternoon with some short crust pastry.

    Looking at the distribution of events and whether they are random come under Poisson Analysis.

    I’m not sure how to apply it to this exercise, but I will give it some thought….

  40. Peyer Kellner in today’s Sunday Times puts DC on 297 with EM on 262. If he is right, and I suspect he may be, it is game over and it is only likely to get worse for Ed & Ed.
    The swingback could never come from Labour, it has to come from UKIP and it is now moving.

  41. @ Unicorn

    you are making very brave claims above. Did you also test the data against air pressure, level of humidity in the air, etc?

    The day of the week is not part of the design, so you can’t introduce it just like this. It is a huge methodological error.

  42. @David in Oxford

    Calling this election this far out, given how tight it is, is a very brave move.

  43. @CMJ

    As the two tests measured different things, you can’t really compare the results.

    No?

    My understanding of your own procedures is that you routinely take steps to set aside distorting variability.

    Correct me if I am wrong, but my recollection is that you opt to opt to work with YouGov data alone rather than using polling series based on data collected by the full range of polling companies. I had assumed that your reason for doing this was to eliminate the known biases and variability introduced by including data from all pollsters. To my mind this is a sensible decision and it allows you to pick up effects that would otherwise be undetectable. (What would your CUSUM analyses look like if you included all companies in the plots?)

    If this does reflect your thinking, then presumably you take the view that the more informative analysis is the one that takes steps to cut out unexplained variation.

    Why then – when I do the same thing – do you suggest that the two analysis are much of a muchness with neither being better than the other?

    I am fairly sure I could predict your response if I repeated one of your CUSUM analyses missing polling data from all the different houses. If – after doing this – I failed to reproduce your Tory rise from last week (or any other pattern) you would just put this failure down to my use of noisy data.

    So, I submit that I am just doing what your routinely do in your own analyses.

  44. The Labour lift towards Thursday may be true but that doesn’t mean it has a cause.

    If you toss a coin a million times statistically you will have 500k each of heads and tails, but it is more likely one will come in at more than that one less.

    Equally although you would expect a fifty fifty chance of a head or tail you will get times when you get ten or twenty of one in a row.

    It might just be random variation and nothing more.

    Always be careful not to pay so much attention trying to figure out what your seeing that you end up seeing something that isn’t there. I remember someone calling it “Drowning in a Martian Canal”

    Peter.

  45. Well he owns You Gov, I think, ? Have they been massively wrong before, maybe? Everyone form Ladbrokes to You Gov are calling DC as the largest number of seats and votes, but the question is whether all other parties are prepared to gang up into some giant anti Tory coalition to prevent him carrying on. Peter Kellner suggests that the reaction to this weeks budget may open up a 4 point lead for DC. We shall see!

  46. Alec – “Only possibly explanation if the phenomena is real would be sample bias based on day of the week. Unlikely.”

    It isn’t THAT unlikely – especially Monday’s poll, which has some fieldwork at the weekend, compared to the others which only have fieldwork on weekdays. I think theoretically its perfectly possible… I’ve just never seen any convincing evidence of it happening in practice.

    Taking this year’s YouGov polls so far, there is some variation by day… taking just those weeks that had the full five polls (to remove any distortion by actual change over time), the average Labour lead was biggest on Thursdays (1.1 points), smallest on Tuesday (-0.5 points). A point and a half difference…

    But it is only 10 data points for each day of the week. Doing the same for the whole of last year’s data (again, taking only whole weeks) the biggest Labour lead was on Mondays (average 3.6 points), the smallest on Tuesdays and Wednesdays (average of 3.3 points). The pattern changes and the differences fade to only a third of a point.

    I can imagine why there might well be an effect from whether fieldwork is done on a weekday or weekend… but I can’t imagine why such a pattern would appear out of the blue when last year it just looked like noise.

  47. @ catmanjeff
    I never bet, but I am very tempted to put the Cons on 300 plus one or two now.

  48. @PC

    It might just be random variation and nothing more.

    It certainly could be. As I said in the original comment there is a one in thirty chance that you would get such an effect merely by chance: not that different from throwing a double six.

    The question is whether you should sit up and take notice when you come across a relatively unexpected event.

    What would you say if the same pattern reappeared in the Oct-Dec YouGov polls?

    More random variation that is not worth looking at?

  49. Anthony,

    “but I can’t imagine why such a pattern would appear out of the blue when last year it just looked like noise”

    Labour voters on Zeo Hours contracts having to work Wednesday’s!!!!

    Peter.

  50. @ Amber

    I was contemplating to build a similar model, but gave it up, because I needed the DK, won’t vote figures, and these aren’t weighted.

    Anyway, it would be a very computation intensive model that you are outlining, because you would need to test against data density. With modern computers it’s not an issue really. There very good Linux programmes with graphic interface that would do the job.

    You probably need very good demographic data (normally available) and party affiliation data (core, peripheral, floating, undecided, unlikely, never) which could be a challenge, but with good assumptions you can use the data to test this proxy measure.

    So it is a nested, conditional probability model. I assume you want to use the polling data to test it, and make it “learn”. It is quite feasible, but I have doubt about the data as I said above.

    You can use a Bayesian model (borrowing maybe one of the econometric ones) using one of the open code softwares for this or a fuzzy set one (based on the method of difference, as you are assuming that identical influences affec individuals and constituencies differently). Berkeley use to have a free software to do the calculations and testing.

1 2 3 4 11